Microsoft bets on AI agents, not apps, and dynamic UI with Project Solara

Project Solara rethinks computing with AI agents replacing apps, dynamic interfaces adapting in real time, and hardware designed around tasks instead of traditional software workflows

Most consumer-facing devices are built around apps. You open one, navigate a fixed interface, and complete a task within it. Microsoft’s Project Solara is built on the assumption that this model is no longer optimal.

 

Unveiled at Build 2026, Solara is a platform designed for what Microsoft calls “agent-first devices,” where AI agents replace traditional applications as the primary way users interact with software. According to Microsoft’s official announcement, these devices are designed to “run AI agents instead of traditional applications,” shifting the core unit of computing from apps to tasks.

 

Unlike a conventional operating system, Solara is structured as a chip-to-cloud platform. It combines an Android Open Source Project (AOSP) base with Microsoft’s enterprise stack, including identity, device management, and Copilot-powered agents, allowing the system to operate across both local hardware and cloud infrastructure.

 

MIcrosoft Project Solara: From app workflows to task execution

In a traditional system, even simple workflows require multiple applications. Scheduling a meeting, summarising documents, and sending follow-ups involves switching between calendar, email, and productivity apps.

  READ: Microsoft launches 'Scout' AI assistant that works like executive assistant

  Solara attempts to remove this fragmentation. According to Microsoft, agents in Solara are designed to interpret user intent and execute tasks across systems without requiring explicit app-level interaction. Instead of opening apps, users issue commands or prompts, and the system orchestrates the workflow in the background.

 

This effectively shifts computing from an application model to an orchestration model, where the system decides:

• which services to call
• how data is retrieved
• and how results are presented

How the system is structured

At a technical level, Solara operates in three layers.

Device layer (AOSP foundation)

The base of the system is built on Android Open Source Project, which handles hardware-level functions such as connectivity, sensors, and device security. However, this is not a standard Android implementation. Microsoft layers its own device ecosystem platform on top, integrating enterprise identity (Entra ID), device management (Intune), and security controls.

Agent and orchestration layer

This is the core of Solara. Agents operate across services rather than within apps. They can access enterprise systems, cloud data, and local inputs, and are coordinated through an orchestration layer that determines how tasks are executed. 

For example, a request such as “prepare a summary of today’s meetings” would involve:

• retrieving calendar data
• accessing documents and notes
• generating summaries using AI models
• structuring output for presentation

All of this happens without exposing underlying applications to the user. 

Interface layer (dynamic output)

The final layer is the interface, which is not fixed.

‘Just-in-time UI’: removing fixed interfaces

Microsoft describes Solara’s interface model as “just-in-time UI”. According to the company, instead of pre-designed layouts, interfaces are generated dynamically based on context, device type, and user intent. 

This means the same task can produce different outputs depending on the device:

• a wearable may show a short summary
• a larger display may show detailed controls and data

The implication is that developers no longer need to design separate interfaces for different screen sizes. Instead, the system generates the interface at runtime.

What Solara-powered devices may look like

Microsoft showcased two reference designs to illustrate how this model could work in practice.

 

The first is a wearable badge, positioned as an always-connected assistant for enterprise workers. It includes a touchscreen, microphone array, camera, and biometric authentication, allowing users to interact with agents through voice, glanceable information, or touch.

 

The second is a desk companion device with a touchscreen and ambient sensors, designed to provide continuous access to AI agents. When connected to an external display, it can also function as a cloud PC client.

  READ: What is Anthropic's Mythos AI model and why India's inclusion matters?

]  Both devices are built around a key constraint: they do not run traditional applications. There is no app store or desktop interface. All interactions are mediated through agents.

 

According to ArsTecnica, early pilots are already underway with companies including Best Buy, CVS Health, Levi’s, and Target.

A broader industry shift

Microsoft is not alone in moving toward agent-first computing, but its approach differs in how deeply it integrates the model into the stack.

 

Google, for instance, is embedding its Gemini AI as the primary interaction layer in Android XR devices, including smart glasses, where interaction is designed to be conversational and context-driven rather than app-based. The company also introduced Gemini Intelligence at its I/O 2026 conference, last month, which essentially builds upon existing hardware such as smartphones and is designed to carry out tasks on behalf of the user across multiple apps and websites. It also plays with the concept of dynamic UI with its customisable widget feature.

 

OpenAI is also working toward AI-first hardware through its partnership with former Apple designer Jony Ive, with reports indicating a focus on devices where interaction is centred around AI agents rather than traditional software interfaces.

 

At the infrastructure level, NVIDIA is preparing for this shift. “AI agents will be the largest users of computing,” said Jensen Huang, the company’s founder and chief executive, while introducing its Vera CPU designed for agentic workloads.

 

What distinguishes Solara is that it attempts to combine all of these elements into a single platform, spanning hardware partnerships with Qualcomm and MediaTek, operating system design, and enterprise deployment.

Constraints and open questions

Despite its scope, Solara remains an early-stage system. The platform is currently limited to concept hardware and pilot environments, and its effectiveness depends heavily on the maturity of AI agents.

 

The model assumes that agents can reliably interpret intent, coordinate across services, and deliver accurate results. In practice, these systems are still evolving, particularly in complex, multi-step workflows.

 

There is also the question of ecosystem transition. The app-based model is deeply embedded, with established developer tools, distribution systems, and business models. Replacing that model would require not just new technology, but a fundamental shift in how software is built and consumed.